A Data Management Platform for Personalised Real-Time Energy
Total Page：16
File Type：pdf, Size：1020Kb
A data management platform for personalised real-time energy feedback David Murray 1, Jing Liao 1, Lina Stankovic 1, Vladimir Stankovic 1, Richard Hauxwell-Baldwin 2, Charlie Wilson 2, Michael Coleman 3, Tom Kane 3, Steven Firth 3 University of Strathclyde 1, University of East Anglia 2, Loughborough University 3 Abstract This paper presents a data collection and energy feedback platform for smart homes to enhance the value of information given by smart energy meter data by providing user-tailored real-time energy consumption feedback and advice that can be easily accessed and acted upon by the household. Our data management platform consists of an SQL server back-end which collects data, namely, aggregate power consumption as well as consumption of major appliances, temperature, humidity, light, and motion data. These data streams allow us to infer information about the household’s appliance usage and domestic activities, which in turn enables meaningful and useful energy feedback. The platform developed has been rolled out in 20 UK households over a period of just over 21 months. As well as the data streams mentioned, qualitative data such as appliance survey, tariff, house construction type and occupancy information are also included. The paper presents a review of publically available smart home datasets and a description of our own smart home set up and monitoring platform. We then provide examples of the types of feedback that can be generated, looking at the suitability of electricity tariffs and appliance specific feedback. 1 Introduction Research into Home Energy Management Systems (HEMS) is currently developing at a rapid rate, spurred on by the imminent requirement of utility suppliers to be able to supply advanced services for improving customer retention and make their business more attractive. This reinforces the push on research into data collection and analysis and the development of energy disaggregation algorithms [1, 2, 3] as well as the development of activity recognition [4, 5] and other decision support tools to provide much needed energy feedback for long-term user engagement and interaction. To facilitate the above research efforts, it is important that researchers have access to data which can be used for the purposes of building HEMS, providing effective energy feedback and studying domestic energy consumption behaviour. There are many datasets publically available that contain raw power consumption readings. Some datasets consist of a small number of houses at very high sampling rates of the order of kHz (e.g., BLUED [6], REDD [1]), and others have a large number of houses with low sampling rates of the order of 2 or 10 minutes (e.g., UK HES [7]). With the UK nationwide rollout of smart meters underway, the UK-based REFIT project aims to provide data similar to what can be expected from UK smart meters, as defined by the Smart Metering Equipment Technical Specification (SMETS) [9] proposal by UK’s DECC (Department of Energy and Climate Change). Therefore we chose an 8-10 second sampling rate for electrical readings, similar to what can be provided by households that install a Consumer Access Device in their homes to read the Smart Meter measurements directly [9]. This is a higher rate than the half-hourly rate that can be provided via opt-in to energy suppliers and their trusted third parties should home owners wish to be given additional feedback in terms of general energy usage. Compared to other experimental datasets, such as REDD, BLUED, which have supplied a number of variables such as reactive/apparent power, energy, frequency, phase angles, voltage and current, our dataset catalogues active power usage as this is the data that will be available to both UK households and utilities. In addition to electricity readings we measure temperature, occupancy and light levels. Our dataset also provides analytical data such as how home automation equipment is used and can incorporate coded qualitative data on domestic activities and technology usage routines. This provides greater interpretative depth compared to similar databases (REDD, BLUED, GREEND [10] etc.) where qualitative data, the occupancy group and building type are not provided so it is harder to draw conclusions on the efficiency of the participants against similar homes. This paper focuses on the design and development of a data management platform for personalised real-time energy monitoring, developed as part of the UK REFIT project [8], and the potential understanding of household consumption and behaviour that can be inferred from this platform from simple to more advanced analytics. The main contribution of the paper is to demonstrate the ease of energy feedback generation as well as a reliable backend for remotely monitoring real-time data such as active power and environmental parameters. The backend includes a database that enables simple and quick access to variety of quantitative data related to domestic energy research via simple queries. To the best of our knowledge, there is no similar data platform publicly available of the same scale or duration which is supplemented by such in-depth participant information. The remainder of the paper is organised as follows. Section 2 provides a review of similar publically available datasets and sets out the motivation for this study. Section 3 explains the setup of our data monitoring platform including details of measurements collected during our study. Section 4 details the software and database backend for data management. Section 5 discusses the initial analysis of the data gathered with a view towards improving the quality of personalised feedback that can be generated for households. Section 6 concludes the paper. 2 Review of Publicly Available Domestic Electrical Consumption Datasets A range of datasets are already available which can be used in a similar way to the REFIT electrical dataset. We review these to help show why our database aims to help supply more in-depth information than existing databases. It can be seen in Table 1 that there are a number of datasets that can be used for advanced analytics such as energy disaggregation. Projects that run for a long period of time (> 1 year) tend to focus on a small number of houses (< 10) (AMPds, IHEPCDS), while short projects, of the order of days or several months, may have a large sample set. Table 1: Publically Available Dataset Comparison Dataset Location Study # of Houses # of Features Resolution of Study Period Appliance Sensors per House ACS-F1 Switzerland 2, 1 hour N/A 100, P, Q, I, f, V, 10 sec [11] sessions (10 types) Φ (2013) AMPds Canada 1 year 1 19 I,V,pf, F, P, 1 min [12] (2012- Q, S 2013) BLUED [6] United 8 days 1 Aggregated I, V, switch 12 kHz States (2011) events GREEND Austria & 1 year 9 9 P 1 Hz [10] Italy (2013 - 2014) HES [7] United 1 month 251 13-51 P 2 min Kingdom (255 houses) 1 year (26 houses) (2010 – 2011) iAWE [13] India 73 days 1 33, V, I, f, P, S, 1 Hz (2013) (10 appliance E, Φ level) IHEPCDS France 4 years 1 3 I, V, P, Q 1 min [14] (2006 – 2010) OCTES Scotland, 4 – 13 33 Aggregated P, Φ, 7 secs 2 [15] Iceland & months Energy Finland (2012 - price 2013) REDD [1] United 3 – 19 days 6 9 - 24 Aggregate: 15 kHz States (2011) V, P; Sub- (aggr.) metered: P 3 secs (sub) REFIT [8] United 2 years 20 11+ P, Energy 8 seconds Kingdom (2013 – price 2015) Smart* United 3 months 1 sub- 25 circuit, P, S 1 Hz [16] States (2012 – metered, + 2 29 appliance (circuit), P 2013) (Aggregated + (sub) Sub-metered) Tracebase Germany N/A (2012) 15 158 P 1 – 10 sec [17] (43 types) UK-DALE United 499 days 4 5 (house 3) Aggregated 16 Khz [18] Kingdom (2012 – 53 (house 1) P, Sub P, (aggr.) 2014) switch 6 secs status (sub) Active Power (P), Reactive Power (Q), Apparent Power (S), Energy (E), Frequency (f), Phase Angle (Φ), Voltage (V) and Current (I). The largest dataset in terms of length and study size is HES [7]. However, the monitoring interval is small and sampling rate of 2 minutes is low. This makes it less suitable for energy disaggregation research as it becomes harder to distinguish individual appliances and events. It does however provide information about the recorded households, including type, size and number of occupants. The OCTES [15] dataset is similar to our own, recording Active Power, phase and energy cost. It has a slightly shorter study duration, a slightly larger study size and data are recorded at a similar resolution to ours. It publically provides the electrical data for each house; however, it does not provide any information about the houses other than their geographical location based on country. The example of analysis given describes the use of a sauna in one household; however, this information is not provided publically, therefore an assumption would have to be made as to the power consumption. Tracebase [17] is a dataset that contains appliance signatures, which helps disaggregation research. The signatures, obtained with 1 second sampling rate, could be used for training but cannot be used to investigate usage patterns as no information is given with respect to the make or model of the appliances being monitored. It currently contains 43 different devices. Each device has a number of recordings from different days and different houses. Also available in the dataset are readings of date, power and average power sampled at 8 seconds. In all the above datasets, there is no information about the environment where consumption is taking place. Our platform provides an appliance survey and household composition which will help researchers to develop a much more in-depth picture and allow for more types of analysis.